An opinionated list of LLM roles, or personas, grouped by topic.
See the "data/roles-grouped-sorted.csv" file for the actual roles.
This is my personal version of a collection of roles, also known as personas, prompt modifiers, custom instructions, or similar terms. These are for human-AI interactions with LLMs, Large Language Models such as ChatGPT.
The roles are stored in a single file and were classified by me into groups like "Fun+Leisure", "IT-Expert", "Business", and more.
Specifying a role works like this:
"As a web developer and software tester, be critical of the code input I provide and provide me with a list of issues in the code."
... but you do this on a terminal. Then you can copy and paste this role description and put it into the LLM's web form. Alternatively, supply this as an additional command-line argument to the LLM.
Processing/Usage is further illustrated in USAGE.md below, and in the shell script find-role.
How I use it in my workflow: See USAGE.md, EXAMPLE.md, and my repo perplexity-api-search.
FOR-AND-AGAINST-PROMPTING.md describes why such roles are becoming less important these days, actually.
CC-0, see LICENSE file.